Implicit and explicit statistical learning of tone sequences across spectral shifts

Abstract

a b s t r a c tWe investigated how the statistical learning of auditory sequences is reflected in neuromagnetic responses in implicit and explicit learning conditions. Complex tones with fundamental frequencies (F0s) in a five-tone equal temperament were generated by a formant synthesizer. The tones were subsequently ordered with the constraint that the probability of the forthcoming tone was statistically defined (80% for one tone; 5% for the other four) by the latest two successive tones (second-order Markov chains). The tone sequence consisted of 500 tones and 250 successive tones with a relative shift of F0s based on the same Markov transitional matrix. In explicit and implicit learning conditions, neuromagnetic responses to the tone sequence were recorded from fourteen right-handed participants. The temporal profiles of the N1m responses to the tones with higher and lower transitional probabilities were compared. In the explicit learning condition, the N1m responses to tones with higher transitional probability were significantly decreased compared with responses to tones with lower transitional probability in the latter half of the 500-tone sequence. Furthermore, this difference was retained even after the F0s were relatively shifted. In the implicit learning condition, N1m responses to tones with higher transitional probability were significantly decreased only for the 250 tones following the relative shift of F0s. The delayed detection of learning effects across the sound-spectral shift in the implicit condition may imply that learning may progress earlier in explicit learning conditions than in implicit learning conditions. The finding that the learning effects were retained across spectral shifts regardless of the learning modality indicates that relative pitch processing may be an essential ability for humans.

Statistical learning of an auditory sequence and reorganization of acquired knowledge: A time course of word segmentation and ordering

“…This suggests that the P2 response resists adaptation more than the N1 response. Due to this long time constant of P2 response modification, in our previous study, when participants listened to auditory sequences for 5 min, statistical learning effects were reflected in the N1 but not the P2 responses (Daikoku et al, 2014(Daikoku et al, , 2015. By training for voice-onset-time perception, the N1-P2 peak-topeak amplitude increased over a long time scale of 10 days, whereas the P1-N1 peak-to-peak amplitude had no significant difference at any time scale (Tremblay et al, 2001).…”

“…In addition, these statistical learning effects reflected in neural responses could be observed in both the ERPs and the magnetic counterparts of ERPs. This neural effect could be observed in the responses at approximately 50 ms (i.e., P1/P1m) (Paraskevopoulos et al, 2012;Daikoku et al, 2016), 100 ms (i.e., N1/N1m) (Abla et al, 2008;Furl et al, 2011;Daikoku et al, 2014Daikoku et al, , 2015Koelsch et al, 2016), and 200 ms (i.e., P2/P2m) (Furl et al, 2011) after stimulus onset.…”

“…Previous studies on auditory statistical learning, using a secondorder Markov chain (Markov, 1971(Markov, , reprinted in 1971Furl et al, 2011;Daikoku et al, 2014Daikoku et al, , 2015 or a word segmentation task (Abla et al, 2008;Paraskevopoulos et al, 2012), have shown that auditory statistical learning can be reflected in neural responses. Previous studies using two different experimental paradigms have shown similar results: neural responses to tones that appear frequently in reference to the most recent tones are more rapidly reduced than responses to tones that appear infrequently.…”

“…Previous neurophysiological studies have investigated only a single aspect of language learning at a time, such as the acquisition of words (Abla et al, 2008;Paraskevopoulos et al, 2012) or grammatical structure (Furl et al, 2011;Daikoku et al, 2014Daikoku et al, , 2015. In our real-world learning environments, when humans listen to language sequences, they are exposed to the linguistic traits of word boundary and word order patterns.…”

“…According to previous studies on statistical learning that used a Markov stochastic model or a word segmentation task, tones that appear with higher transitional probability (i.e., more-predictable tones) lead to reduced amplitude and shortened neural response latencies, whereas tones that appear with lower transitional probability (i.e., less-predictable tones) preserve the neural response amplitude (Abla et al, 2008;Furl et al, 2011;Paraskevopoulos et al, 2012;Daikoku et al, 2014Daikoku et al, , 2015. In the present study, using both the Markov stochastic model and a word segmentation task, we hypothesized that the amplitudes for the final tone within words would be significantly reduced compared to those for the initial tone, after learning word borders (i.e., word segmentation), and that the amplitudes for words that appeared with higher transitional probability would be significantly reduced compared with those with lower transitional probability.…”